Anemia Severity Detection in Pediatric Patients Through REXlayer-Integrated Deep Learning and Eye Conjunctival Imaging

Anemia in young children presents significant health challenges, requiring accurate and non-intrusive diagnostic methods. This study introduces the Recurrent Expansion Layer (REXlayer), a novel component for learning the relationships between independent and dependent variables in pretrained neural networks. The REXlayer is integrated with a single-layer Long Short-Term Memory (LSTM) network to detect anemia severity (non-anemic, mild, moderate, severe) from non-intrusive eye conjunctival images. A sophisticated preprocessing pipeline, including image segmentation, feature extraction, scaling, feature selection, and class balancing, ensures high-quality input data. The REXlayer within the LSTM framework processes these preprocessed images to predict anemia severity levels. Experimental results show that the REX-LSTM model outperforms a standard LSTM, demonstrating enhanced accuracy and reliability in anemia detection. This approach advances pediatric anemia diagnosis and highlights the effectiveness of combining advanced preprocessing techniques with innovative deep learning layers for medical image analysis, offering a less invasive option for young patients.